Enabling Heterogeneous Performance Analysis for Scientific Workloads

TL;DR

This paper discusses the development of Adaptyst, an open-source, architecture-agnostic performance analysis tool for scientific workloads on heterogeneous systems, exploring eBPF-based methods to enhance performance insights.

Contribution

It introduces Adaptyst, a novel framework for performance analysis across heterogeneous architectures, and evaluates eBPF-based methods for future integration.

Findings

eBPF-based methods like Uprobes and USDT are promising for performance analysis

Performance analysis complexity varies across different architectures

Roadmap for integrating eBPF techniques into Adaptyst

Abstract

Heterogeneous computing integrates diverse processing elements, such as CPUs, GPUs, and FPGAs, within a single system, aiming to leverage the strengths of each architecture to optimize performance and energy consumption. In this context, efficient performance analysis plays a critical role in determining the most suitable platform for dispatching tasks, ensuring that workloads are allocated to the processing units where they can execute most effectively. Adaptyst is a novel ongoing effort at CERN, with the aim to develop an open-source, architecture-agnostic performance analysis for scientific workloads. This study explores the performance and implementation complexity of two built-in eBPF-based methods such as Uprobes and USDT, with the aim of outlining a roadmap for future integration into Adaptyst and advancing toward heterogeneous performance analysis capabilities.